Metabolic Remodelling in the Mouse Embryo and Placenta at Mid-Gestation

2. Results

2.1. Stage-Associated Metabolic Remodelling in the Mouse Embryo from E9.5 to E13.5

To characterise the metabolic landscape of the post-implantation mouse embryo, we collected embryos at five consecutive developmental stages—E9.5, E10.5, E11.5, E12.5, and E13.5—and subjected them to untargeted LC–MS-based metabolomics (Figure 1A). After stringent quality filtering, 639 endogenous metabolites were retained for downstream analysis (Figure S1B). Inter-replicate correlation within each stage was high, confirming the robustness of the dataset (Figure S1C).

Principal component analysis (PCA) revealed that samples formed a continuous trajectory along PC1 according to developmental time (Figure 1B). E9.5 and E10.5 samples clustered closely, as did E12.5 and E13.5 samples, whereas the largest separation was observed between E11.5 and E12.5. This pattern indicates that the embryonic metabolome undergoes progressive remodelling, with the most pronounced compositional shift occurring immediately after E11.5.

Pairwise comparisons between consecutive stages were consistent with this view. The largest numbers of significantly altered metabolites were observed in the E11.5 versus E10.5 and E12.5 versus E11.5 comparisons, marking the period around E11.5 as the most metabolically dynamic phase of the examined window (Figure 1C-G). Among the top altered metabolites, octanoylcarnitine increased from E9.5 to E10.5, 5-thymidylic acid increased from E10.5 to E11.5, and fructose-1,6-bisphosphate increased from E11.5 to E12.5 (Figure 1C–F). In addition, several phosphatidylinositol and phosphatidylglycerol species were among the metabolites that increased between E12.5 and E13.5. KEGG pathway enrichment analysis of the E11.5 versus E10.5 comparison identified enrichment of fructose and mannose metabolism, taurine and hypotaurine metabolism, and arginine biosynthesis (Figure S1G). The E12.5 versus E11.5 comparison was enriched for glycine, serine and threonine metabolism, the pentose phosphate pathway, and valine, leucine and isoleucine biosynthesis (Figure S1H). These pathways are broadly involved in carbon utilisation, amino acid interconversion, and biomass production, suggesting that the embryo adjusts its metabolic network as organogenesis progresses.

To identify metabolites whose abundance increased across the full developmental window, we applied one-way ANOVA followed by a directional criterion requiring higher abundance at E13.5 than at E9.5. This approach identified a substantial set of metabolites with progressively increasing abundance (Figure 1H). Mfuzz clustering of all detected metabolites further revealed distinct temporal patterns, including clusters characterised by continuous increase, continuous decrease, and transient changes (Figure 1I). The increasing clusters contained metabolites such as myo-inositol(Placidi et al., 2024), taurine(Guérin et al., 2001), and several nucleotide derivatives, which have established roles in signalling, protection against reactive oxygen species, and nucleotide metabolism, respectively.

Taken together, these analyses establish that the embryonic metabolome undergoes stage-associated remodelling, with the most dynamic changes centred around E11.5. This metabolic reorganisation coincides with a period of rapid organ expansion and increasing biosynthetic demand, supporting the view that E11.5 represents a key developmental window in which metabolic and morphological changes are tightly coordinated.

2.2. Metabolomic Comparison of the Placenta and Embryo Reveals a Shared Set of Metabolites Increasing from E10.5 to E12.5

Because the placenta becomes the dominant site of maternal–fetal exchange during this window, we reasoned that metabolites increasing in both the embryo and placenta might represent factors whose availability becomes increasingly important for development. We therefore performed untargeted metabolomics on placentas collected at E10.5, E11.5, and E12.5, matching the embryonic time points (Figure 2A). PCA of placental metabolomic profiles showed a gradual shift along PC1 with developmental stage, indicating progressive metabolic maturation of the placenta itself (Figure 2A).

Pairwise comparisons identified 82 upregulated metabolites in E11.5 versus E10.5 and 51 in E12.5 versus E11.5 (Figure 2B–D). Among the top metabolites, cholic acid increased from E10.5 to E11.5, whereas maltotriose and maltotetraose increased from E11.5 to E12.5. These examples illustrate that the placenta undergoes metabolic changes involving both bile acid-related and carbohydrate-related pathways during this interval. To focus on metabolites that increased in both tissues, we compared the ANOVA-defined increasing metabolites from the embryo and placenta. This analysis identified 73 shared metabolites whose abundance increased with developmental time in both tissues (Figure 2E). Heatmap visualisation confirmed coordinated increases in these metabolites across matched embryonic and placental samples (Figure 2F). This shared set likely contains both metabolites that are transferred from mother to fetus and metabolites that are synthesised independently in each tissue, reflecting the functional interplay between placental supply and embryonic demand.

KEGG pathway enrichment analysis of the 73 shared metabolites revealed significant enrichment of pantothenate and CoA biosynthesis, ascorbate and aldarate metabolism, and taurine and hypotaurine metabolism (Figure 2G). Pantothenate is an essential precursor of coenzyme A, which is required for fatty acid oxidation, the TCA cycle, and numerous biosynthetic reactions(Barritt et al., 2024), while taurine functions as an osmolyte and antioxidant(Guérin et al., 2001). Representative metabolites from the shared set, including D-fructose and pseudouridine, showed progressive increases in both tissues from E10.5 to E12.5 (Figure 2H–I).

Collectively, these results identify a set of metabolites whose abundance rises in both the embryo and placenta during mid-gestation, suggesting coordinated metabolic adaptation between the two tissues. This shared set provides a focused starting point for identifying factors that may support embryonic growth during this critical period.

2.3. WGCNA Identifies Development-Associated Metabolite Modules and Converges on a Core Set of Metabolites Co-Increasing in the Embryo and Placenta

The analyses above identified metabolites that increased in the embryo or in both embryo and placenta. However, metabolites rarely act in isolation; they are organised into coordinated networks. To capture this organisation and further prioritise candidates, we performed WGCNA on the embryonic metabolomic dataset. This analysis grouped metabolites into modules of co-abundant compounds with similar accumulation patterns (Figure 3A). Among the identified modules, Module 3 showed the strongest positive correlation with developmental stage, followed by Module 2, whereas Module 1 displayed the strongest negative correlation (Figure 3B). This indicates that the embryonic metabolome can be decomposed into distinct modules that are differentially associated with developmental progression.

Pathway enrichment analysis of Module 3, the module most strongly associated with development, revealed enrichment of ascorbate and aldarate metabolism, pantothenate and CoA biosynthesis, and valine, leucine and isoleucine biosynthesis (Figure 3C). Module 2 was enriched for fructose and mannose metabolism, the pentose phosphate pathway, and taurine and hypotaurine metabolism (Figure 3D). These pathways converge on central carbon metabolism, nucleotide precursor generation, and antioxidant defence, all of which are likely to be in high demand during rapid organogenesis. The distinct enrichment profiles of Modules 2 and 3 also suggest that developmentally associated metabolites are organised into multiple regulatory layers, rather than a single linear programme.

To derive a high-confidence set of candidate metabolites, we intersected the union of Module 2 and Module 3 with the 73 metabolites that increased in both embryo and placenta (Figure 3E). This convergence yielded 56 core metabolites that were simultaneously developmentally associated in the embryonic WGCNA network and co-increasing in the two tissues. Correlation network analysis of the top 30 core metabolites revealed densely connected groups (Figure 3F), suggesting that these metabolites may be co-regulated or functionally interrelated. Several representative core metabolites—glyceric acid, arginylglycine, riboflavin, myo-inositol, taurine, and ergothioneine—displayed progressive increases in both embryonic and placental samples (Figure 3G–L). These compounds include a riboflavin derivative involved in cofactor synthesis(Jiménez-Nava et al., 2026), the signalling lipid precursor myo-inositol(Placidi et al., 2024), the antioxidant ergothioneine(Chen et al., 2024), and taurine, all of which have potential roles in supporting development. Glyceric acid is connected to serine/glycine metabolism and one-carbon flux, while arginylglycine represents a dipeptide whose accumulation may reflect increased amino acid turnover.

In summary, by combining tissue-level time-course analysis with network-based module detection, we narrowed the candidate space to 56 core metabolites that show coordinated increases in the embryo and placenta and are embedded in development-associated networks. This core set provides a rational basis for prioritising metabolites with potential functional relevance to mid-gestation development.

2.4. Integrative Single-Cell and Metabolomic Analysis Reveals a Metabolic Programme Associated with Definitive Erythroid Maturation

Because whole-embryo metabolomics cannot resolve cell-type-specific contributions, we next integrated our metabolomic data with a published single-cell transcriptomic atlas of the E9.5–E13.5 mouse embryo(Cao et al., 2019). We re-embedded the atlas using UMAP and annotated 38 cell types spanning major lineages and organ primordia (Figure 4A). Cell-type proportion analysis revealed substantial shifts in cellular composition across the five stages (Figure 4B), consistent with rapid tissue expansion and lineage diversification during organogenesis.

To identify metabolic programmes that are specific to individual cell types, we focused on cell-type-enriched metabolic genes. Among the 38 annotated populations, the definitive erythroid lineage showed enrichment of several metabolic pathways, including phenylalanine metabolism, beta-alanine metabolism, and nicotinate and nicotinamide metabolism (Figure 4C–D). Nicotinate and nicotinamide metabolism is of particular interest because it feeds NAD+ biosynthesis, which is essential for redox balance, energy metabolism, and numerous enzymatic reactions. The enrichment of this pathway in definitive erythroid cells suggests that NAD+ metabolism may be especially important during erythroid maturation.

To examine how metabolic gene expression changes during erythroid maturation, we inferred pseudotime for definitive erythroid lineage cells using Palantir and visualised the differentiation trajectory with CellRank (Figure 4E–F). Along this trajectory, early erythroid progenitors expressed higher levels of Sulf1, Kcnb2, Kcnq5, and Kdm5b, whereas more mature cells progressively upregulated Nmnat2, Cacna2d1, Grik3, Slc2a1, and Snca (Figure 4G). Among the late-stage genes, Nmnat2 encodes a key enzyme in NAD+ biosynthesis, directly linking the transcriptional programme to the enriched nicotinate/nicotinamide pathway(Ryu et al., 2018). Slc2a1 encodes GLUT1, the primary glucose transporter in erythrocytes(Martí-Mateos et al., 2026), which supports glycolytic energy production in maturing red cells. The upregulation of these genes suggests that definitive erythroid maturation is accompanied by an increased capacity for NAD+ synthesis and glucose uptake.

To determine whether these transcriptional changes were reflected at the metabolite level, we examined the activity of the nicotinate and nicotinamide metabolism pathway using embryonic metabolomic data. Pathway activity increased progressively from E10.5 to E13.5 (Figure 4H), and the contributing metabolites—niacinamide, NAD, and 1-methylnicotinamide—showed corresponding increases during this period (Figure 4I). The coordinated rise in pathway activity and metabolite abundance, together with the transcriptional upregulation of Nmnat2 in maturing erythroid cells, supports the idea that enhanced NAD+ metabolism is a feature of definitive erythroid maturation during this developmental window.

Taken together, these results identify a metabolic programme associated with definitive erythroid maturation, characterised by increased nicotinate/nicotinamide metabolism and NAD+ biosynthesis. Because erythroid cells are essential for oxygen transport and the establishment of functional circulation, this programme may contribute to the maturation of the haematopoietic system during organogenesis.

2.5. Integrative Regulon–Metabolite Correlation Analysis Reveals Coordinated Variation Between Transcriptional Programmes and Metabolic Modules

The coordinated behaviour of development-associated metabolite modules raised the question of whether these metabolic changes are accompanied by corresponding transcriptional programmes. To explore this, we performed pySCENIC analysis to infer regulon activity from the single-cell transcriptomic data. The top cell-type-specific regulons included known lineage regulators, such as Gata6(+) in cardiomyocytes, TCF3(+) in the definitive erythroid lineage, and Spi1(+) in white blood cells (Figure 5A). Spi1 encodes PU.1, a master regulator of haematopoietic specification, and its identification as the top regulon in the white blood cell compartment likely reflects the inclusion of haematopoietic stem/progenitor cells within this broad cluster.

We next correlated the activity of these regulons with the abundance of metabolites in the two development-associated WGCNA modules. For Module 3, high positive correlations were observed for Thra(+), Maf(+), Arnt(+), Spi1(+), and Atf7(+) (Figure 5B). Thra encodes the thyroid hormone receptor α, which is involved in energy metabolism and cardiovascular development; Arnt is a central mediator of hypoxia-inducible responses(Mandl and Depping, 2014); and Spi1 is essential for haematopoietic lineage specification(Oikawa et al., 1999). For Module 2, high positive correlations were found for Arnt(+), Nr1i2(+), Pparg(+), Ikzf2(+), and Gfi1(+) (Figure 5C). Among these, Pparg is a nuclear receptor with established roles in lipid metabolism, while Gfi1 is identified as one direct target of RUNX1 and critical regulators of hematopoietic transition(van der Meer et al., 2010). The enrichment of haematopoiesis-related regulons among the top correlates of both modules is consistent with the erythroid-focused findings from the single-cell analysis and points to coordinated variation between haematopoietic transcriptional programmes and these metabolic modules.

Representative metabolite–regulon correlations—glyceric acid with Spi1(+), glyceric acid with Gfi1(+), pantothenic acid with Maf(+), and arginylglycine with Arnt(+)—further illustrate these associations (Figure 5D–G). Glyceric acid is connected to serine/glycine metabolism and one-carbon flux; pantothenic acid is the precursor of CoA; and arginylglycine reflects amino acid turnover. The positive correlations between these metabolites and haematopoiesis-related regulons indicate coordinated changes between specific metabolic outputs and transcriptional programmes during this window.

Finally, to provide a functional context for future testing of candidate metabolites, we established a rotating-vessel post-implantation embryo culture system adapted from Hanna and colleagues (Figure 5H). Embryos explanted at E7.5 and cultured ex vivo developed to approximately E11.5, recapitulating key morphological milestones including somitogenesis, neural tube closure, and limb bud formation (Figure 5I). In this system, we also performed a preliminary assessment of glutamine metabolism: exogenous glutamine was taken up by cultured embryos, and pharmacological inhibition of glutaminase with CB839 impaired embryo development and gastruloid formation (Figure S2). These observations provide a starting point for using this culture platform to test candidate metabolites identified in our analysis.

Collectively, these analyses reveal coordinated variation between development-associated metabolic modules and specific transcriptional programmes, particularly those related to haematopoietic development, and establish an ex vivo platform for future functional interrogation of candidate metabolites.

3. Discussion

In this study, we combined untargeted metabolomics of the mouse embryo and placenta with single-cell transcriptomic analysis to characterise the metabolic changes that accompany organogenesis from E9.5 to E13.5. We identified a set of metabolites whose abundance increases coordinately in the embryo and placenta, prioritised 56 core candidates through network-based analysis, and linked these metabolic changes to transcriptional programmes associated with definitive erythroid maturation. We further established an ex vivo embryo culture platform that can be used to test the functional relevance of candidate metabolites.

The metabolic remodelling we observed was progressive, with the most pronounced changes occurring around E11.5. This stage represents a pivotal point in mouse development, when organ primordia are being established and the embryo transitions from relatively simple to increasingly complex body plans. Previous metabolomic analysis of mid-gestation mouse embryos identified E10.5–E11.5 as a major metabolic transition(Perez-Ramirez et al., 2024; Solmonson et al., 2022), and our data are broadly consistent with that finding. In our dataset, the largest number of differentially abundant metabolites was detected between E11.5 and E12.5, followed by E11.5 versus E10.5. These results indicate that the period immediately around E11.5, rather than a single exact time point, is the most metabolically dynamic phase of the examined window. Slight differences in the exact interval of maximum change may reflect differences in embryo staging, sample processing, or metabolite coverage; however, the overall conclusion that E11.5 lies at the centre of a major metabolic reorganisation is consistent between studies. This metabolic reorganisation coincides with the functional maturation of the chorioallantoic placenta and the shift from yolk sac-derived to maternal blood-borne nutrition(Kalisch-Smith, 2025), as well as with rapid expansion of organ primordia and the onset of definitive haematopoiesis(Chen et al., 2022; de Bruijn et al., 2000). The convergence of these developmental events supports the notion that E11.5 represents a critical window in which metabolic and developmental programmes become tightly coordinated.

This study included matched placental metabolomic data, as the placenta became the primary route of nutrient delivery during this window; metabolites that increase in both tissues may represent factors whose availability is coupled to embryonic demand. The 73 shared increasing metabolites and the 56 core metabolites that also mapped to development-associated WGCNA modules provide a focused set of candidates for future functional testing. Among these, myo-inositol and taurine are particularly notable. Myo-inositol is a precursor for phosphoinositide signalling and has established roles in neural tube closure(Greene and Copp, 1997), while taurine functions as an osmolyte and antioxidant with known importance for cardiovascular and neural development(Lambert et al., 2015). The presence of pantothenate-related metabolites is also of interest, given that coenzyme A is required for fatty acid oxidation and a wide range of biosynthetic pathways(Barritt et al., 2024). These observations are consistent with a high demand for membrane biogenesis, signalling lipids, and cofactor synthesis during mid-gestation organogenesis.

The integration of our metabolomic data with single-cell transcriptomics highlighted the definitive erythroid lineage as a site of active metabolic specialisation. Erythroid cells showed enrichment of nicotinate and nicotinamide metabolism at the transcriptional level, accompanied by progressive increases in niacinamide, NAD, and 1-methylnicotinamide in the embryo. The upregulation of Nmnat2, which encodes a key NAD+ biosynthetic enzyme, in maturing erythroid cells further supports the coordination of these observations. NAD+ is essential for redox balance and energy metabolism, and erythroid cells face particular demands in this regard as they accumulate haemoglobin and prepare for oxygen transport. The establishment of functional circulation is a prerequisite for efficient nutrient distribution within the growing embryo, and metabolic support of erythroid maturation may therefore contribute to this process. These findings point to a functional link between embryonic metabolism and the maturation of the haematopoietic system.

Our regulon analysis identified several transcriptional programmes whose activity correlated with development-associated metabolic modules. Among the top correlates were regulons linked to haematopoietic specification, such as Spi1 and Gfi1, and regulators of metabolic adaptation, such as Arnt and Pparg. These observations reveal coordinated variation between haematopoietic transcriptional programmes and metabolic outputs during this developmental window, although the causal relationships underlying these associations remain to be established.

Yet, this study has several limitations. The metabolomic measurements were performed on whole-embryo and whole-placenta lysates and therefore represent tissue-level averages, which cannot resolve cell-type-specific or spatially localised metabolic differences. In addition, the functional relevance of the candidate metabolites identified here remains to be tested directly.

The core metabolite set defined in this study provides a concrete basis for functional follow-up. Using the ex vivo embryo culture platform described here, we will next test whether supplementation with selected candidate metabolites can support development beyond the current E11.5 barrier. Given the association we observed between metabolic changes and erythroid maturation, a particularly promising direction is to determine whether promoting haematopoietic development through specific metabolites improves nutrient transport and thereby extends embryo culture. Ultimately, the metabolomic atlas and candidate list presented here provide a foundation for systematically improving post-implantation embryo culture and for understanding how metabolism supports organogenesis.

4. Methods

4.1. Mouse Embryo Collection

All animal procedures were approved by the Animal Research Committee guidelines of Zhejiang University (research license ZJU20230458) and performed in accordance with relevant ethical guidelines. C57BL/6J mice were maintained under specific-pathogen-free conditions on a 12-h light/dark cycle with ad libitum access to food and water. Timed matings were set up overnight, and the morning of vaginal plug detection was designated as embryonic day 0.5 (E0.5). Pregnant females were euthanised by cervical dislocation, and embryos were dissected at E9.5, E10.5, E11.5, E12.5 and E13.5 in ice-cold phosphate-buffered saline (PBS). Extra-embryonic membranes were removed, and individual embryos were snap-frozen in liquid nitrogen within 5 min of dissection and stored at −80 °C until metabolomic extraction. Five biological replicates were collected for E9.5, E10.5, and E11.5, and four biological replicates for E12.5 and E13.5. Placentas were collected at E10.5, E11.5, and E12.5, with four biological replicates per stage. All embryo and placenta samples weighed at least 40 mg. To ensure consistent sample loading for LC–MS, frozen samples were first ground and homogenised, and an aliquot equivalent to approximately 20 mg of tissue was used for metabolite extraction. All samples were processed and analysed in a single batch to avoid potential batch effects.

4.2. Rat Serum Preparation

Rat serum was prepared accordingly(Li et al., 2025). Adult male Sprague-Dawley rats (8–9 weeks old, 280–350 g) maintained under specific pathogen-free conditions were fasted for 18–24 h with access to water to normalise blood glucose. Under 2.5%–4% isoflurane anaesthesia, blood was drawn from the abdominal aorta using a 21G needle attached to a 20 ml syringe, then immediately transferred to 15 ml centrifuge tubes and spun at 4,000 g for 15 min at 4 °C. To promote fibrin clot formation, the tubes were left at 4 °C for an additional 1.5–2 h. The resulting clot was gently compressed with curved forceps to release residual serum, which was then collected into fresh 15 ml tubes. Serum was heat-inactivated in a 55 °C water bath for 45 min and passed through a 0.22 μm filter. Aliquots were stored at −80 °C, and each batch was tested for its ability to support embryo development prior to experimental use.

4.3. Post-Implantation Embryo Culture

Embryos were cultured using the rotating-vessel system essentially as described by Hanna and colleagues(Aguilera-Castrejon et al., 2021). Briefly, E7.5 embryos (late gastrulation stage) were dissected from decidua in pre-warmed DMEM medium supplemented with 10% (vol/vol) fetal bovine serum, 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin. The collected embryos were cultured in DMEM/F12 medium supplemented with 75% rat serum, glucose, 1× glutamine, 11 mM HEPES, and 1× antibiotic-antimycotic solution. Glucose and oxygen concentrations were increased daily in a stepwise manner. From embryonic day (E) 7.5 (designated as day 0), the glucose concentration was initiated at 3 mg/mL and increased by 0.5 mg/mL per day. Oxygen tension was progressively elevated from 5% to 13%, 18%, and 21% on consecutive days starting from E7.5, and then maintained at 21% thereafter. Cultures were maintained at 37 °C under a pressure of 6.5 psi. Medium was refreshed every 24 h. For inhibitor experiments, CB839 was dissolved in DMSO and added to the culture medium at a final concentration of 200 μM from day 1 of culture; an equivalent volume of DMSO was used as a vehicle control. Embryo development was monitored daily by bright-field microscopy, and the developmental stage was assessed by established morphological criteria.

4.4. Immunofluorescence

Embryos were fixed in 4% paraformaldehyde in PBS overnight at 4 °C, permeabilised with 0.1% Triton X-100 in PBS for 30 min, and blocked with 10% normal donkey serum in PBS for 1 h at room temperature(Li et al., 2025). Primary antibodies used were anti-Sox2 (1:200). After overnight incubation at 4 °C, embryos were washed and incubated with Alexa Fluor-conjugated secondary antibodies (1:500) for 1 h at room temperature. Nuclei were counterstained with DAPI. Images were acquired using a confocal microscope.

4.5. Gastruloid Formation and Treatment

Mouse embryonic stem cells were maintained on mitotically inactivated mouse embryonic fibroblasts in standard ES cell medium. To generate gastruloids, cells were dissociated and aggregated in 96-well ultra-low attachment plates according to established protocols(Beccari et al., 2018). Approximately 300 stem cells per well were seeded in N2B27 medium and allowed to aggregate into embryoid bodies (EBs). At 48 hours post-seeding, the cultures were treated with 3 µM CHIR99021 (CHIR). After 24 hours of treatment, the medium was replaced with fresh N2B27 medium to withdraw CHIR, and subsequently refreshed daily. After 96 h, well-formed gastruloids were treated with CB839 (200 μM) or vehicle, and morphology was assessed at 120 h. Gastruloid length was measured using ImageJ.

4.6. Metabolite Extraction for LC-MS

Embryos and placentas were snap-frozen in liquid nitrogen immediately after dissection. Each sample (approximately 40 mg) was ground and homogenised, and an aliquot corresponding to approximately 20 mg of tissue was used for extraction. The aliquot was homogenised in 200 μL of water using a bead-beater, and 800 μL of methanol/acetonitrile (1:1, v/v) was added for metabolite extraction. The mixture was centrifuged at 14,000 g for 20 min at 4 °C, and the supernatant was dried in a vacuum centrifuge. For LC–MS analysis, the dried residue was reconstituted in 100 μL of acetonitrile/water (1:1, v/v), centrifuged at 14,000 g for 15 min at 4 °C, and the supernatant was used for injection. A pooled quality control (QC) sample was prepared by mixing equal aliquots of all reconstituted samples and was injected periodically to monitor instrument stability. All embryo and placenta samples were analysed in the same batch.

4.7. Liquid Chromatography–Mass Spectrometry

Untargeted metabolomics was performed using an UHPLC system (1290 Infinity LC, Agilent Technologies) coupled to a quadrupole time-of-flight mass spectrometer (TripleTOF 6600, AB Sciex) at Shanghai Applied Protein Technology Co., Ltd. For HILIC separation, an ACQUITY UPLC BEH Amide column (2.1 mm × 100 mm, 1.7 μm; Waters) was used. In both positive and negative electrospray ionisation (ESI) modes, the mobile phase consisted of A = 25 mM ammonium acetate and 25 mM ammonium hydroxide in water, and B = acetonitrile. The gradient was: 95% B for 0.5 min, linearly reduced to 65% B in 6.5 min, then reduced to 40% B in 1 min and held for 1 min, followed by an increase to 95% B in 0.1 min, with a 3-min re-equilibration period. The ESI source conditions were: Ion Source Gas 1 60, Ion Source Gas 2 60, curtain gas 30, source temperature 600 °C, and IonSpray Voltage Floating ± 5500 V. In MS-only acquisition, spectra were acquired over m/z 60–1,000 Da with an accumulation time of 0.20 s/spectrum. In auto MS/MS acquisition, product ion scans were acquired over m/z 25–1,000 Da with an accumulation time of 0.05 s/spectrum, using information-dependent acquisition (IDA) in high-sensitivity mode. The collision energy was 35 V ± 15 eV, and declustering potential was 60 V (positive) and −60 V (negative). Isotopes within 4 Da were excluded, and up to 10 candidate ions were monitored per cycle.

4.8. Metabolomics Data Processing

Raw MS data were converted to mzXML format using ProteoWizard MSConvert and processed with XCMS software. Peak detection was performed using the centWave algorithm (m/z tolerance 10 ppm, peakwidth c(10, 60), prefilter c(10, 100)). Peak grouping was performed with bw = 5, mzwid = 0.025, and minfrac = 0.5. CAMERA was used for annotation of isotopes and adducts. Only features with more than 50% nonzero measurements in at least one sample group were retained. Metabolite identification was performed by matching accurate mass (m/z error < 10 ppm) and MS/MS spectra against an in-house database constructed with authentic standards, as well as public databases including HMDB, KEGG, and METLIN.

4.9. Preprocessing and Quality Control

Data matrices from positive and negative modes were separately processed. Features with >50% missing values were removed, and values ≤0 were set to NA. Quantile-based probabilistic quotient normalisation (PQN) was applied, followed by QC-based filtering (features with relative standard deviation >30% in QC samples were discarded). Remaining missing values were imputed with half the minimum positive value per feature. Data were log₂-transformed, and the positive- and negative-mode matrices were merged by taking the maximum intensity per metabolite. QC samples were then removed from the final matrix. Metabolite names were standardised, and redundant features sharing the same HMDB ID were collapsed by retaining the feature with the highest detection score. Endogenous metabolites were retained based on HMDB origin annotation or SuperClass. The final dataset comprised 639 metabolites across 35 biological samples. Between-replicate reproducibility was assessed by Pearson correlation.

4.10. Differential Metabolite Analysis

Pairwise comparisons between consecutive developmental stages (E10.5 vs E9.5, E11.5 vs E10.5, E12.5 vs E11.5, E13.5 vs E12.5) were performed using the limma package. Metabolites were considered significantly altered when they met the following criteria: variable importance in projection (VIP) > 1 from a PLS-DA model fitted to all stages, log₂ fold change > 0.32, and P < 0.05.

4.11. Multi-Stage One-Way ANOVA and Heatmap

One-way ANOVA was performed across the five stages using limma’s F-test with empirical Bayes moderation. Metabolites with adjusted P < 0.05 were considered significant. The top 100 most significant metabolites were selected for heatmap visualisation. Rows were ordered by the difference in mean expression between early (E9.5 + E10.5) and late (E12.5 + E13.5) stages.

4.12. Mfuzz Temporal Clustering

Mean abundance per stage was calculated for each metabolite (n = 4 replicates). Data were standardised (mean = 0, standard deviation = 1) and analysed using the Mfuzz package(Futschik and Carlisle, 2005). The number of clusters was set to 6, and the fuzzifier parameter was estimated automatically. The cluster showing a monotonic increase across stages (largest linear regression slope) was designated as the “continuously increasing” cluster.

4.13. Weighted Gene Co-Expression Network Analysis (WGCNA)

WGCNA was performed on all 693 metabolites using the WGCNA package(Zhang and Horvath, 2005). A signed network was constructed with a soft-thresholding power chosen to achieve scale-free topology (R² > 0.8). Modules were detected using dynamic tree cutting with a minimum module size of 15 and a merge cut height of 0.25. Module–trait correlations were calculated between module eigengenes and developmental stage. The module with the highest absolute correlation was selected for further analysis. Hub metabolites within the module were defined by intramodular connectivity.

4.14. Pathway Enrichment Analysis

Pathway enrichment was performed using the MetaboAnalyst web server (Pathway Analysis module) with the Mus musculus SMPDB library. For each metabolite set of interest (e.g., WGCNA module, Mfuzz cluster, intersecting metabolites), the list of metabolite names was submitted. Results were visualised as bubble plots or bar charts in R.

4.15. Pathway Activity Scoring

Single-sample gene set enrichment analysis (ssGSEA) was used to estimate pathway activity scores(Barbie et al., 2009; Subramanian et al., 2005). Gene sets (pathway definitions) were obtained from the SMPDB database by matching target pathway names to SMPDB pathway identifiers, and metabolite members were mapped to detected metabolites via standardised HMDB IDs. For each sample, ssGSEA scores were calculated using the GSVA package with a Gaussian kernel. Activity trajectories were plotted as mean ± s.e.m. across replicates for each stage.

4.16. Single-Cell Transcriptomic Data Analysis

Published single-cell RNA-seq data of E9.5–E13.5 mouse embryos were obtained(Cao et al., 2019). Data were processed using the Scanpy pipeline (version 1.2.2)(Wolf et al., 2018). After quality filtering to remove low-quality cells and potential doublets, counts were normalised and log-transformed, and 3000 highly variable genes were identified. Principal component analysis was performed with 30 PCs, and a neighbourhood graph was constructed using n_neighbors = 50, followed by UMAP embedding for visualisation. Cell types were annotated based on canonical marker genes and the original annotations provided by the source study. In total, 38 cell types were retained for downstream analysis. Cell-type proportions across developmental stages were calculated by counting annotated cells per stage and normalising to the total cell number per stage.

To identify cell-type-enriched metabolic genes, differential expression analysis was performed between each cell type and all other cells using the Wilcoxon rank-sum test implemented in Scanpy. Genes with significant enrichment and a metabolic annotation were retained. Dotplots were generated to visualise the expression of selected metabolic marker genes across cell types, with dot size representing the fraction of expressing cells and colour intensity indicating scaled average expression. KEGG pathway enrichment analysis of cell-type-enriched metabolic genes was performed using GSEApy with the KEGG gene set library(Fang et al., 2023).

4.17. Pseudotime and Differentiation Trajectory Analysis of Definitive Erythroid Lineage

To characterise the developmental progression of definitive erythroid lineage cells, we first subset cells annotated as definitive erythroid lineage from the full atlas. Pseudotime was inferred using Palantir (version 1.4.5)(Setty et al., 2019), which estimates differentiation trajectories from single-cell expression data. The resulting pseudotime ordering was used as input for CellRank (version 2.0.7) to visualise differentiation trajectories(Weiler and Theis, 2026). Arrows in the resulting trajectory plots indicate the inferred direction of differentiation from early progenitors toward more mature erythroid cells. Gene expression dynamics along pseudotime were visualised for selected metabolic genes by fitting smoothed expression trends and plotting scaled expression values against pseudotime.

4.18. pySCENIC Regulon Analysis

To infer cell-type-specific transcriptional regulatory programmes, we performed pySCENIC analysis (version 0.12.1)(Aibar et al., 2017; Kumar et al., 2021). Approximately 50,000 cells were sampled across developmental stages and cell types to balance computational efficiency and representation. The analysis consisted of three steps: (i) co-expression module detection using GRNBoost2, (ii) cis-regulatory motif enrichment using RcisTarget, and (iii) regulon activity scoring using AUCell. Regulon specificity scores (RSS) were calculated to identify regulons that are preferentially active in specific cell types. The top cell-type-specific regulon for each of the 38 cell types was selected based on RSS and visualised as a dotplot.

4.19. Correlation Analysis Between Metabolites and Regulons

To examine the relationship between development-associated metabolic modules and transcriptional regulatory programmes, we correlated metabolite abundance with regulon activity scores. WGCNA module metabolites were correlated with regulon activity across samples using Pearson correlation. Regulons showing the highest positive correlations with each module were identified and reported. In addition, representative core metabolites from the intersection set were correlated with selected regulons of interest. Correlation strength and direction were visualised using scatter plots, with each point representing a biological sample.

4.20. Statistical Analysis and Reproducibility

All metabolomic analyses were performed in R (version 4.4.3). For boxplots, the box limits represent the interquartile range (IQR), the centre line denotes the median, and whiskers extend to the most extreme data point within 1.5 × IQR from the box. Individual replicates are shown as points. No statistical methods were used to pre-determine sample size; sample sizes were chosen based on feasibility and prior experience with embryonic metabolomics. Experiments were not randomised, and investigators were not blinded to allocation during outcome assessment, as morphological assessment of embryo development was unambiguous.

4.21. Data and Code Availability

The code and data supporting the findings of this study are available in an anonymous GitHub repository for peer-review purposes at the following link: https://anonymous.4open.science/r/Metabolomics-mouse-embryo-and-placenta-try4F3C.

The repository includes all input and processed data, as well as R and Python scripts required to reproduce the analyses. The single-cell transcriptomic data used in this study were obtained from Cao et al. (2019) and are available in the Gene Expression Omnibus (GEO) database under accession number GSE119945. All other data supporting the findings of this study are available from the corresponding author upon reasonable request.